A systematic review of the evidence on decoupling of GDP, resource use and GHG emissions, part II: synthesizing the insights

Strategies toward ambitious climate targets usually rely on the concept of ‘decoupling’; that is, they aim at promoting economic growth while reducing the use of natural resources and GHG emissions. GDP growth coinciding with absolute reductions in emissions or resource use is denoted as ‘absolute decoupling’, as opposed to ‘relative decoupling’, where resource use or emissions increase less so than does GDP. Based on the bibliometric mapping in part I (Wiedenhofer et al, Environ. Res. Lett. 15 063002), we synthesize the evidence emerging from the selected 835 peer-reviewed articles. We evaluate empirical studies of decoupling related to final/useful energy, exergy, use of material resources, as well as CO2 and total GHG emissions. We find that relative decoupling is frequent for material use as well as GHG and CO2 emissions but not for useful exergy, a quality-based measure of energy use. Primary energy can be decoupled from GDP largely to the extent to which the conversion of primary energy to useful exergy is improved. Examples of absolute long-term decoupling are rare, but recently some industrialized countries have decoupled GDP from both production- and, weaklier, consumption-based CO2 emissions. We analyze policies or strategies in the decoupling literature by classifying them into three groups: (1) Green growth, if sufficient reductions of resource use or emissions were deemed possible without altering the growth trajectory. (2) Degrowth, if reductions of resource use or emissions were given priority over GDP growth. (3) Others, e.g. if the role of energy for GDP growth was analyzed without reference to climate change mitigation. We conclude that large rapid absolute reductions of resource use and GHG emissions cannot be achieved through observed decoupling rates, hence decoupling needs to be complemented by sufficiency-oriented strategies and strict enforcement of absolute reduction targets. More research is needed on interdependencies between wellbeing, resources and emissions.


Introduction
Many policy documents and scientific publications, including those of the IPCC, assume that economic growth will continue to be a cornerstone of thriving future societies. However, if economic growth is accompanied by increases of resource use and emissions (Steinberger et al 2013, Hickel andKallis 2019), it may threaten chances of meeting future sustainability transformation goals. Achieving targets such as the SDGs (TWI2050, 2018 or the Paris climate accord to limit global heating to 1.5 • C-2.0 • C (IPCC 2018) requires reducing emissions of greenhouse gases (GHG) to zero around 2050, and most likely also absolute reductions of the use of natural resources such as energy or materials in many world regions. In many scenarios, net negative emissions, achieved either through reforestation and other landbased 'natural climate solutions ' (Griscom et al 2017) or negative emission technologies (Fuss et al 2018, Nemet et al 2018, Rogelj et al 2019, are required after 2050 to bring the climate back from an overshoot over the climate-change mitigation targets to the specified target level. The need for 'negative emissions' emerges in all scenarios that fail to achieve sufficient cuts in emissions in the first half of the century (IPCC 2018).
If achieving ambitious climate and sustainability targets should be reconciled with continued GDP growth, an absolute decoupling (or 'de-linking'; (Vehmas et al 2003)) of GDP from the use of biophysical resources and/or emissions is a logical necessity (UNEP 2011a, Hickel and Kallis 2019, Jackson and Victor 2019, Parrique et al 2019, UNEP-IRP 2019. In this set of two articles, we present a systematic review of the empirical literature on past (de)coupling of resource use and emissions and GDP. Part I has provided a bibliometric mapping of this literature and focuses on how decoupling is empirically analyzed in various strands of research (Wiedenhofer et al 2020). Here in part II, we synthesize the evidence in this literature with respect to observed historical (de)coupling and discuss its implications for science and policy.
We analyze the scientific literature on the relationships between economic output (most commonly measured as inflation-corrected GDP) and resource use or emissions and the observed rates of relative and absolute decoupling. We aim at elucidating the potential contribution of past and ongoing gains in economy-wide efficiency and productivity towards absolute decoupling and zero carbon futures. The socio-ecological systems perspective of socio-economic metabolism (Fischer-Kowalski 1998, Pauliuk and Müller 2014, Pauliuk and Hertwich 2015, Haberl et al 2019 stresses that socio-economic systems continuously require materials and energy for all economic activity and the reproduction of humans, livestock, and all manufactured capital, which necessarily leads to emissions and waste. From this perspective, materials, energy, waste and emissions are inextricably interlinked and therefore need to be treated jointly, an idea sometimes denoted as 'resource nexus' (Bleischwitz et al 2018b). The broad scope of this systematic review was motivated by the aim to capture such systemic linkages, as they are increasingly acknowledged as important for both science and policy (Haberl et al 2019). The scale and patterns of socio-economic metabolism are also directly entangled with past and future development pathways, as well as with socioeconomic structures and policies. To capture such linkages, and to address the question to what extent the resource/GDP relations might be amenable to active intervention, the review also aims to map the key strategies discussed by the literature to achieve decoupling (section 4).
It is important to distinguish resource decoupling (e.g. decoupling of GDP from energy or material use) from impact decoupling (e.g. the decoupling of GDP from GHG emissions) (UNEP 2011a, Jackson and Victor 2019). While reduction of resource use will-ceteris paribus-always reduce impacts because fewer resources need to be extracted, processed or disposed of, some (probably not all) impacts can also be reduced and redirected through technological measures (e.g. flue gas treatment or substitution of low-carbon fuels for high-C fuels such as coal or oil products), even if resource use is not reduced. For GHG emissions, such options are intensively researched and may gain importance in the future (based on carbon capture and sequestration or CCS technologies; (Fuss et al 2014)). However, they are currently not deployed and hence are not included in this review, which only covers studies of observed past decoupling, and excludes all model-based studies on future scenarios. This focus is supported by IPCC reports demonstrating that energy efficiency and demand-side measures have less risks and are more benevolent to societies than technological fixes (IPCC, 2014, Creutzig et al 2016. A key issue for decoupling and decarbonization, which plays a big role in this review, is global trade and its role in connecting producers and consumers. There are three complimentary perspectives (Steininger et al 2015). (1) The production-based (territory-based) perspective accounts for resources used in or emissions emerging from a territory. It underlies emission accounts of the UNFCCC.
(2) The consumption-based perspective accounts for resources used or emissions emerging-no matter where in the world-along supply chains and required to meet the final demand of a national economy. Such a perspective is required to account for displacements and problem shifting through international trade, e.g. 'improvements' of energy intensity (energy/GDP) resulting from increasing imports of embodied energy in imported goods that help reducing the need to produce these goods domestically Vuille 2018, Moreau et al 2019). (3) The income-based perspective accounts for resources used in or emissions emerging in the generation of income for a given country (Rodrigues et al 2006, Marques et al 2012. However, the difference between consumption-,production-and income based accounts cannot simply be interpreted as 'leakage' or 'outsourcing' (Jakob and Marschinski 2013), as the attribution of responsibility along supply chains is complex (Rodrigues et al 2006, Rodrigues and Domingos 2008, Steininger et al 2016. Recognition of this challenge has resulted in proposals of various methods to derive displacement indicators (Kander et al 2015, Jiborn et al 2018. Data allowing the allocation of resource use or emissions directly or indirectly occurring along international supply chains to final consumers are recently becoming available through the development of multi-regional input-output models (Peters 2008, Rodrigues et al 2010, Wiedmann et al 2015, Domingos et al 2016, Steininger et al 2015, 2016, Liang et al 2017. The production-, consumption-and income-based perspectives on resource use and emissions can result in widely diverging, if not opposing, results when analyzing the relations between resources/emissions and GDP hence both production-and consumption-based will be considered for a better assessment (see section 5; figure 2). We do not include the income-based perspective because studies with empirical results at the national or global level are rare (Rodrigues et al 2010, Marques et al 2012, Steininger et al 2016, Liang et al 2017 In this evidence synthesis, we consider production-and consumption-based perspectives but restrict ourselves to national-and international studies, acknowledging that substantial amounts of work have been published on sub-national and citylevel decoupling, as well as sectoral-or raw material/energy carrier specific perspectives. Including these literatures would not have been consistent with the comprehensive focus of this review. Moreover, studies with a narrow geographical or thematic scope cannot provide the top-down perspective necessary to identify problem-shifting and rebound effects in the global system in which we are particularly interested. Specifically, we address the following research questions: • What is the empirical evidence for relative or absolute decoupling of economic output from resource use and emissions at the national-to-global level? • Which strategies and policy recommendations are discussed by the literature empirically investigating efficiency and decoupling trends? Do they point towards a 'degrowth' or 'green growth' perspective? • What can be learned from past decoupling trends for achieving future absolute reductions in resource use and GHG emissions?

Methods
In this article, we conduct an evidence synthesis for a body of the 835 peer-reviewed journal articles and book chapters identified in part I . There, we describe a search query to SCOPUS as well as ISI Web of Knowledge and an expert solicitation, yielding 11 609 references covering the time span between the first captured study from January 1972 until June 7, 2019. 8455 articles remained after duplicate removal, which we screened first at the level of titles and abstracts and second at the full-text level, eliminating all non-relevant articles and yielding the final 835 papers for in-depth review. Part I describes these procedures in detail, including criteria for exclusion as well as those applied at the coding stage. It also presents a bibliometric mapping of this body of literature and comparatively discusses the development of the identified research streams and their approaches to investigating decoupling phenomena. For part II (this paper), we proceeded as follows. Because the body of literature on primary energy, territorial CO 2 and on the causality relations between energy use and GDP is very large and recent reviews exist, we relied on these reviews and handpicked references to summarize their implications for the overall topic of this article (section 3.1). We then present an in-depth analysis of the following streams of literature: (1) Studies on useful energy and exergy, and a part of the literature on final energy (section 3.2). (2) Studies on aggregate material and energy flows following a social metabolism approach (section 3.3).
(3) Studies on total GHG emissions as well as studies on carbon emissions from fossil fuel combustion and industrial processes, excluding studies only dealing with territorial CO 2 emissions (section 3.4).
In section 4, we focus on discussing the strategies adopted (explicitly or implicitly) in the empirical decoupling literature. Due to the scope of this systematic review, conceptually and theoretically oriented papers explicitly focusing on policy choices were mostly excluded by the search query. Therefore, our analysis is restricted to policy recommendations and strategies found in papers that have a focus on biophysical evidence rather than politics. For the qualitative mapping and synthesis of strategies and policy recommendations, we drew a random subsample of 15% from the 835 articles, yielding 125 articles for further qualitative content synthesis. We used widely accepted definitions of green growth and degrowth to interpretatively map the 125 papers according to these definitions: • For green growth, we refer to three major international institutions (OECD, UNEP and the World Bank) that promote green growth (OECD 2011, UNEP 2011b, World Bank 2012. Their definitions range from relative decoupling (World Bank 2012) to absolute decoupling (OECD 2011, UNEP 2011b, p 2011, World Bank 2012. Articles were classified as 'green growth' if their framing aimed at absolute or relative decoupling without impeding economic growth. • Articles were classified as 'degrowth' if their framing explicitly challenged the primacy of economic growth over the (absolute) reduction of resource use and emissions, or articles that were agnostic towards economic growth (van den Bergh and Kallis 2012). We included articles in this category, based on their empirical findings, if they at least challenged economic growth as a 'taken for granted' variable. That is, we included articles that either proposed an 'equitable downscaling of economic production and consumption' (degrowth; quote on p 910) or adopted an 'indifferent' (p 912) position towards the effects of certain policy measures on economic growth (a-growth) (van den Bergh and Kallis 2012). • Papers not meeting the above criteria were classified as 'others' . This category mostly includes papers which were primarily concerned with the causality between GDP and energy use or GHG emissions without expressing any aim of reducing emissions or resource use.
We openly coded the subsample (based on abstract, introduction, conclusion, and, if applicable, policy recommendations) according to the strategies and policies they recommended. In a next step, we merged these open codes to derive manageable and meaningful findings. For example, we merged the recommendations 'internalization of external environmental goods' , 'regulate prices' and 'environmental taxes' into the category 'pricing' .

Synthesis of key insights and quantitative evidence on decoupling
In this section, we comparatively review the literature on the relation between economic growth and various resource-use and emission indicators, covering both production-and consumption-based studies. We critically examine the state and trajectory of these research streams and summarize their key insights and quantitative results on relative and absolute decoupling.
We start by summarizing the evidence on the coupling between GDP and primary energy respectively territorial CO 2 emissions, which are closely related because burning fossil fuels (which account for a large fraction of primary energy in most countries) is the dominant source of CO 2 emissions (section 3.1). In contrast to sections 3.2-3.4, this section does not undertake an analysis of all articles within this category; we instead rely on recent major reviews and selected studies. We then summarize the findings on the extent of decoupling between GDP and final energy as well as exergy (section 3.2), i.e. indicators that are much more closely linked to the actual functions, utility and services of energy for socioeconomic activities (Lovins 1979, Haas et al 2008, Kalt et al 2019. Section 3.3 presents the evidence on the (de)coupling between GDP and comprehensive measures of social metabolism derived with the harmonized and internationally applied economy-wide material and energy flow analysis (MEFA) framework (Haberl et al 2004, Fischer-Kowalski et al 2011, Krausmann et al 2017a. This comprehensive perspective covers combustible energy carriers such as fossil fuels, as well as non-metallic minerals, ores and metals and biomass, which are all required for socio-economic activities and are highly interlinked , Krausmann et al 2017a, Bleischwitz et al 2018b. Section 3.4 summarizes the evidence on the coupling between GDP and emissions based on full GHG accounts (including agriculture, forestry, and other land use (AFOLU) and non-carbon greenhouse gases, consumption-based CO 2 emissions as well as territorial and consumption-based full GHG accounts).

Primary energy and territorial CO 2 emissions
Although neo-classical economic growth models (see Aghion and Howitt 2009) do not include energy as a production factor, the relationship of energy use and economic growth has gained significant attention in recent research. Recognizing that standard regression methods are insufficient with regard to avoiding spurious correlation, 10 cointegration and Granger causality tests have been the predominant approaches for time-series statistical analysis from the 1970s onwards (Stern 2011). Cointegration testing identifies long-term equilibria between two or more non-stationary variables (Enders 2014). Granger causality tests analyze the direction of causality, i.e. whether one time series is useful in forecasting another (Granger 1969).
Using these well-established methods, this large body of literature finds that long-run primary energy-GDP cointegration exists across a wide range of temporal and geographic scales. However, the direction of the energy-GDP Granger causality is inconclusive, as directionalities differed according to the considered regions, timeframes and methods used (Ozturk 2010, Stern 2011, Kalimeris et al 2014, Omri 2014, Tiba and Omri 2017. Besides the lack of directionality, energy-GDP Granger causality testing itself is somewhat controversial. For example, Bruns et al (2013) suggest there is a prevalence of model misspecification and publication bias. 11 Other scholars criticize the 'speculative and exploratory' nature of the Granger causality debate (Beaudreau 2010) and that the same methodological approaches continue to be applied although they have proven to be inadequate for resolving the question of directionality (Karanfil 2009, Ozturk 2010, Kalimeris et al 2014, Tiba and Omri 2017. Stern (Stern 1997(Stern , 2011 argues that regardless of whether econometric approaches find empirical 10 Spurious correlation is where variables trending over time appear to be correlated with each other simply because of the shared directionality, but there is no true underlying relationship (Stern 2011). 11 The 'tendency of authors and journals to preferentially publish statistically significant or theory-conforming results' (Bruns et al 2013). evidence for causality in one or another direction, energy is always an essential factor of production. This viewpoint is corroborated by several studies reviewed in section 3.2 and has long been voiced by 'biophysical economists' (Hall et al 1986, Cleveland 1987, Kümmel 2011. Based on a synthesis of energybased and mainstream models of economic growth, (Stern 2011) finds that energy scarcity imposes a strong constraint on economic growth. He also identifies factors that could affect the linkages between energy use and economic output, and are therefore key to gauging the extent of a possible decoupling of GDP from energy use: substitution between energy and other inputs such as capital and labor, technological change, and shifts in the composition of energy inputs and in the economic structure.
Around 80% of global GHG emissions originate from combustion of fossil fuels. Given the historical coupling between primary energy and GDP, we might expect a similar coupling relationship between territorial CO 2 emissions and GDP at the global level (Bassetti et al 2013, Stern 2017. The empirical evidence supports that assertion: global GDP (constant $US2010) grew at 3.5%/year from 1960-2014, while CO 2 emissions grew at 2.5%/year on average (World Bank 2019a); i.e. globally there is relative but no absolute decoupling. Between 2000 and 2014, the relationship was even tighter, as both CO 2 emissions and GDP (constant $US2010) grew at 2.8%/year on average.
At the international level, studies examining the relationships between territorial CO 2 emissions and GDP typically also find weak or relative decoupling (Vollebergh et al 2009, Longhofer and Jorgenson 2017. A few studies find absolute decoupling (Azam and Khan 2016, Roinioti and Koroneos 2017, Chen et al 2018, Madaleno and Moutinho 2018, but these are usually relatively small, short-term reductions of CO 2 emissions (Li et al 2007). A few country-level GDP-CO 2 studies find empirical support for an Environmental Kuznets Curve (EKC) type relationship, whereby CO 2 /capita rises and then falls with rising GDP/capita, i.e. income (Stern 2017). National-level studies (Peters and Hertwich 2008, Kander et al 2015, Azam and Khan 2016, Hardt et al 2018, Moreau and Vuille 2018, Moreau et al 2019, Wood et al 2019a emphasize the role of 'offshoring' emissions (e.g. related to imported goods) and changes in economic structure (e.g. shrinking carbon-intensive industry, larger contributions from service sectors) in distorting the GDP-CO 2 relationship in one or the other direction. Variability in primary energy composition and different stages in renewable energy deployment are also seen as key reasons for differing results regarding the existence of an EKC for CO 2 (Chien and Hu 2007, Fang 2011, Menegaki 2011, Tiwari 2011

Final and useful energy, as well as exergy
Socioeconomic energy flow analyses trace the flow from primary energy extracted from the environment (e.g. crude oil or solar radiation) to final energy put to use in production or consumption (e.g. gasoline or electricity) to useful energy actually performing a specific function (e.g. mechanical work or heat). While data on primary and final energy are readily available from statistical sources in reasonably standardized manner (IFIAS 1974, IPCC 2014, data on useful energy (i.e. the energy actually performing useful work) must be inferred and are only exceptionally reported. Exergy evaluates the thermodynamic quality of these energy flows by quantifying the maximum amount of work (mechanical energy) that a given amount of energy can provide. For example, as electricity can be completely converted into work (i.e. it is equivalent to mechanical work), 1 kWh of electricity has an exergy of 1 kWh. By contrast, the exergy of 1 kWh of heat at 80ºC in an environment at 20ºC is only 0.17 kWh. Data on exergy are not reported by statistical bodies, therefore the community interested in the relation between exergy and economic activity needs to calculate exergy equivalents of primary, final or useful energy flows (Ayres et al 2003).
Research on the relationship between final energy and economic growth is often motivated by questions on energy efficiency. Energy efficiency is usually defined as GDP per unit energy used (see . Some studies find strong linkages between final energy use and GDP (e.g. Kim 1984, Stjepanović 2018, while others find evidence for some degree of decoupling, mostly at the national scale (e.g. Jakob et al 2012, Liddle 2012, Mulder and de Groot 2012, Naqvi and Zwickl 2017. Several studies argue that the observed decoupling can be attributed to structural changes in the economy and outsourcing of energy-intensive activities (e.g. Moreau et al 2019). A recent scenario suggests that low primary energy demand is compatible with staying well below 2 • C and providing services that enable wellbeing for all (Grubler et al 2018) .
Regarding the wealth of studies investigating the energy-GDP relationship applying cointegration and causality tests based on primary energy consumption (see section 3.1), it is somewhat surprising that there are hardly any studies applying such methods to final energy or exergy and GDP. Among the few exceptions are Antonakakis et al (2017) and Belke et al (2011). Both find evidence for bi-directional causality, i.e. for final energy consumption being a driver for GDP as well as vice versa.
The number of studies analyzing exergy flows is comparatively small (see table 1b). Most studies investigating exergy flows find relative decoupling of GDP from primary and final exergy (e.g. Ayres et al 2003, Serrenho et al 2014, Guevara et al 2016, Jadhao et al 2017. In contrast, no significant improvements in intensities or long-term decoupling were found for useful exergy. Some studies even found increasing useful exergy intensities, in particular during periods in which the contribution of industry to GDP respectively industry's share in final energy use rise (e.g. Warr et al 2008, Guevara et al 2016; others did not detect a clear trend (e.g. Serrenho et al 2014Serrenho et al , 2016. Exergy studies found considerable gains in the conversion efficiency from primary to useful exergy (exergy efficiency), but also a slowdown of efficiency gains since the 1970s (Ayres et al 2003. Several macro-economic models use (useful) exergy in addition to capital and labor as factors of production (Warr et al 2008, Warr and Ayres 2012, Santos et al 2018, Sakai et al 2019; these models can generally explain past GDP growth very well, without resorting to residual factors such as autonomous technological growth (Ayres andAyres 2012). This would explain the strong longterm coupling between useful exergy and GDP. Seen from that perspective, the decoupling of primary or final energy/exergy and GDP can be interpreted as an 'economic growth engine' under conditions of scarce resources (Ayres andWarr 2009, Sakai et al 2019). Raising the conversion efficiency of primary to final exergy or final to useful exergy then results in relative decoupling for the former properties while the ratio of useful exergy to growth does not improve substantially-in other words, increases in conversion efficiency drive GDP growth rather than reducing energy use (Ayres andWarr 2009, Sakai et al 2019).

Comprehensive measures of material and energy flows
Studies analysed in this section are based on the social metabolism concept (Fischer-Kowalski 1998); i.e. are studies that comprehensively trace flows of biomass, mineral resources, fossil fuels and many other materials respectively energy sources . In addition to fossil fuels used for the supply of technical energy, biomass used as food and feed also constitutes an important part of a society's energy metabolism (Haberl 2001). Material decoupling is also sometimes denoted as dematerialization (Bernardini and Galli 1993, Cleveland and Ruth 1998, Schandl and Turner 2009. We find very few dematerialization studies prior to the 1990 s (table 2). As also discussed in part I, many of these studies are concerned with compiling MEFA data (MEFA is an extension of MFA that consistently accounts for material and energy flows; see part I) rather than with advanced statistical or econometric analyses, and only 11 econometric dematerialization studies are in our sample of 835 articles.
Long time series of harmonized MEFA data now enable researchers to analyse the interplay between political-economic and material development of countries. Especially at the national level, this analysis commonly analyse how trajectories of material use relate to major phases of socioeconomic or political development, including incisive political events such as the dissolution of the Soviet Union (Krausmann et al 2016) or China's admittance to the World Trade Organisation (Velasco-Fernández et al 2015). At the country level, decomposition analyses (Muñoz and Hubacek 2008, Wenzlik et al 2015, Plank et al 2018 have identified economic growth (of absolute or per capita GDP and/or monetary final demand) as the most important driver of consumption-based measures of resource consumption. (Yu et al 2013) identified technological progress as the most important driver for China, while other drivers were found to have no significant impact on resource use (e.g. Rezny et al 2019 for innovation). The links between GDP growth and material use are also the subject of global studies, covering either aggregated world regions (Behrens et al 2007, Schaffartzik et al 2014 or representative large (> 100) samples of countries (e.g. Steinberger and Krausmann 2011, Steinberger et al 2013, Pothen 2017. At the global scale, a period of relative decoupling after the 1970 s was followed by a period starting 2000 in which global material use accelerated at a similar pace as GDP (Krausmann et al 2018). While many of the studies analyzed in this section apply productionbased accounting principles, a substantial and rising fraction analyze resource flows from a consumptionbased (or 'material footprint'; Wiedmann et al 2015) perspective.
From country case studies based on simple data description to advanced statistical analyses of global samples, relative decoupling has been identified mainly for regions or countries with intermediate economic growth (e.g. USA, European countries) or in countries that experienced socio-economic and political turmoil with corresponding restructuring of their economies Hak 2007, Raupova et al 2014). Absolute reductions of material flows are generally only found in periods of very low economic growth or even recession (Steinberger and Krausmann 2011, Shao et al 2017, Wu et al 2019. Accelerated industrialization and high rates of economic growth, as observable in China in the last decades, often coincide with a growth of material use matching or even outstripping economic growth (Xu and Zhang 2007). The post-World War II boom in the world's wealthiest economies is not widely analysed, with most studies relying on data that does not reach further back than 1970. Hence there is little opportunity to compare the rapid growth phase in the 1950 s found by long-term studies (e.g. Krausmann Table 1. Analysis of the studies on final energy, useful energy and exergy. All studies with one exception reported in the last column refer to production-based (territorial) accounting principles; very few report on the difference between the growth rate of GDP and resource use, so these columns were omitted. Where available, quantitative information on decoupling was integrated in the text in the last column. Acronyms: APEC… Asia-Pacific economic cooperation; DEA…Data envelopment analysis; EU…European Union; IEA…International Energy Agency; EU-KLEMS…Capital (K), labour (L), energy (E), materials (M) and service (S) inputs database of the EU; GHG…Greenhouse gas; ICT…Information and communication technology; LINEX…Linear-exponential production function; NUTS… Nomenclature des unités territoriales statistiques; OLS…Ordinary least square analysis; STAN…STructural analysis database of the OECD; TPES…Total primary energy supply; TFEC…Total final energy consumption; UK…United Kingdom; USA…United States of America.    times faster than DMC/cap early on, growth rates declined thereafter. Material productivity (GDP/DMC) grew fast in stagnation phase (1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991) and collapse phase (1992)(1993)(1994)(1995)(1996)(1997)(1998).         At the same time, it appears that reductions or stagnation in the use of the domestic resource base is often associated with rising importance of trade. In contrast to those measures of decoupling based on territorial indicators, consumption-based perspectives unveil a reversal of trends with efficiencies deteriorating instead of improving and no evidence even for relative decoupling (Giljum et al 2014a, Pothen andSchymura 2015, Wiedmann et al 2015). The integrated, more holistic perspective achieved by considering trade-offs over longer periods as well as across spatial scales is important in assessing the possibilities of and necessary conditions for any future (relative or absolute) decoupling. Currently, decoupling appears to depend on prior use and accumulation of materials and on extractive expansion and rising material flows elsewhere. As long asthis is the case, decoupling cannot be achieved in the long-term or universally.

(De)coupling GDP from total GHG emissions
Reporting of territorial CO 2 emissions from fossil fuel combustion and industrial processes such as cement manufacture is rather straightforward because these emissions can be calculated stoichiometrically from fuel use respectively cement production data. These emissions have been reported for a long time, and are readily available from sources such as CDIAC (Carbon Dioxide Information Analysis Center, https://cdiac.ess-dive.lbl.gov/) for many countries and the global total. Hence, there is a large literature on the decoupling of GDP from territorial CO 2 emissions (section 3.1). By contrast, full GHG accounts also need to quantify emissions from land-use and land-cover changes (LULUCF) as well as highly uncertain and strongly context-dependent emissions such as those of CH 4 and N 2 O. The quantification of 'carbon' respectively GHG footprints (i.e. consumption-based accounts of carbon or GHG emissions) started a bit over a decade ago (Peters and Hertwich 2008, Hertwich and Peters 2009, Peters et al 2011, Lenzen et al 2013, 12 and up to now these studies generally include only fossil-fuel and industrialprocess related emissions, whereas LULUCF emissions of carbon (i.e. changes of the carbon balance of ecosystems resulting from land use, land-use change 12 These studies were not found by the search query as they lacked keywords filtered by the query. We cross-checked elasticities between GHG footprints and GDP as reported in these studies (where available), which confirmed the results of the literature analyzed in Table 3. or forestry) are not systematically accounted for in these databases.
Five studies (Lozano and Gutiérrez 2008, Valadkhani et al 2016, Bampatsou et al 2017, Beltran-Esteve and Picazo-Tadeo 2017 use Data Envelopment Analysis techniques, a method providing efficiency rankings of countries, which show that most countries could reduce their emissions if catching up with the most efficient ones, but does not directly deliver insights on decoupling. Studies searching for an EKC often find no indication for the existence of a turning point (Li et al 2007, Koirala et al 2011, not even a large-scale study of 129 countries (Sanchez and Stern 2016) as well as a global study (Fernandez-Amador et al 2017). A study of 27 EU countries found differently shaped EKCs, but only four countries with an inverted U shape (Jesus Lopez-Menendez et al 2014). A study on Australia 1970-2007 found some evidence for an EKC related to energy, and a declining trend for GHG per GDP . Another study predicts an EKC for Russia (Yang et al 2017), another an EKC for CH 4 for Sub-Saharan Africa (Zaman et al 2017). Overall, however, there is little support for the inverted U-shape hypothesis.
A considerable number of studies used descriptive trend analyses, generally finding relative decoupling, for example for the OECD 1970-2001(Guillet 2010, the Czech Republic (Solilová and Nerudová 2015) and China (Cohen et al 2019). A study of OECD countries covering 1999-2012 found that GHG emissions were constant while GDP grew on considerably (Gupta 2015). A study for Greece (Angelis-Dimakis et al 2012) found that GHG emissions were highest around the year 2000 and then declined somewhat. Decomposition analyses generally find GDP to be an upward driver of GHG emissions. For example, Duarte et al (2013) find that GDPinduced demand growth overwhelmed technologyinduced GHG emission reductions in 11 industrialized countries 1995-2005; similar results were reported for the Baltics (Streimikiene and Balezentis 2016). Xu et al (2014) show that in China 1996-2011, GDP growth was the most important driver of rising emissions. By contrast, from 1999-2009 the EU-27 overall slightly reduced energy use and CO 2 emissions through structural change and improved energy/CO 2 efficiency; GDP growth counteracted but not annihilated these efficiency improvements (Cruz and Dias 2016). In Australia, total GHG emissions have been slightly reduced, whereas industrial CO 2 emissions continued to increase, which was achieved by reductions in LULUCF/agricultural emissions (Leal et  Footprint studies often find that territory-based emissions grow more slowly or even fall while consumption-based emissions increase (e.g. UK 1992-2004, see Baiocchi and Minx 2010;global: Simas et al 2017). There are, however, necessarily also countries where the situation is reversed, e.g. Norway 1980-2000(Faehn and Bruvoll 2009. In 29 highincome countries for the period 1991-2008, GDP was found to drive both territorial and consumptionbased emissions; relative decoupling existed for territorial but not for consumption-based CO 2 (Knight and Schor 2014).
Decoupling was found to be insufficient for reaching climate targets in a study of 120 countries for 2005-2015 (Fanning and O'Neill 2019). Absolute decoupling is found in a footprint-study of GHGs for Sweden 2008-2014 (Palm et al 2019). Most noteworthy is a study of 18 countries with declining CO 2 emissions (both consumption and productionbased) that is discussed in more detail in section 5 ( Le  Quéré et al 2019). Overall, the studies summarized in table 3 suggest that very recently, absolute decoupling between GDP and CO 2 or GHG emissions can be found in some countries, but even in those cases decoupling is so far insufficient to address stringent climate targets, and it is driven by policies promoting renewable energy and energy efficiency (Le Quéré et al 2019).

Strategies for decoupling-green growth versus degrowth
In order to elucidate the perspective on economic growth adopted in empirical decoupling studies, we assessed a random sub-sample of 15% of the 835 articles in terms of their political or strategic assumptions and/or conclusions, as visible in their introduction and conclusions sections respectively the policy-recommendations given (if available). Due to the search query, this body of literature contained only quantitative, empirical studies of decoupling and excluded qualitative policy analyses. Hence almost none of the 125 selected articles focused primarily on strategies or policies for a zero-carbon society and the strategic conclusions or policy recommendations drawn from the quantitative analyses are often rather formulaic. 31% of the articles mentioned no strategies or policy recommendations at all, while 69% provided policy recommendations or strategic conclusions in varying detail.
With regard to their overall framing and aims, 64% of the analyzed articles followed a green growth perspective, that is, they aimed at analyzing absolute or relative decoupling in a given period and territory, and provided policy recommendations in this direction. In line with the literature, a green growth perspective is mainly concerned with 'making growth processes resource-efficient' (Hallegatte et al 2011, p 2) and 'stimulating demand for green technologies, goods, and services' (OECD 2011, p 5), but presents economic growth (measured as increase of GDP) as a set variable. Interestingly, this framing was also common in articles that did not find empirical evidence for absolute decoupling, implying that these studies at least implicitly valued continuation of GDP growth higher than achieving set environmental goals. Only 3% of the articles adopted a degrowth perspective and were open to question the primacy of economic growth. These 'degrowth' studies usually did not explicitly argue in favor of reducing GDP growth; they rather questioned to what extent it would be possible to sustain GDP growth when aiming to reduce resource use or emissions and might hence be classified as 'growth agnostic' , i.e. a-growth (van den Bergh and Kallis 2012). A striking number of one third of the analyzed literature was concerned only with the correlation or causality between energy or resource use and economic growth without explicitly addressing the challenge of decoupling or decarbonization. Policy recommendation in this literature, if at all given, follow a standard green growth repertoire. Some studies which found that growth in energy use Granger-causes GDP growth even argued that saving energy should be viewed cautiously as a policy goal, as it could threaten GDP growth (Yu 2012, Belloumi andAlshehry 2015). Figure 1 summarizes the strategies and policy recommendations given in the articles according to their frequency. Most interestingly, although many articles conclude that absolute decoupling is empirically rarely found, the recommendations to a large extent stick to a green growth repertoire of increasing efficiency, promoting renewable energy and introducing technological solutions and market-based mechanisms (e.g. internalizing or increasing environmental costs through pricing, attract foreign direct investments, financialization or emission trading). Many articles furthermore call for a restructuring of the economy that turns from fossil-energy intensive industrial production towards the service sector. The figure also shows that policy recommendations hardly contain any 'demand-side measures' (not even environmental awareness). Absolute reductions of resource use and emissions (as opposed to relative improvements) are mentioned in <2% of the subsample.
The analysis shows that the large majority of this literature does not question the GDP growth paradigm, even if the empirical evidence suggests that it contradicts officially committed climate policy goals. Policy recommendations point towards a standard repertoire (i.e. efficiency, technology, innovation) that is not further discussed or questioned. Given the focus of the review on studies that quantitatively analyze the relationship between resource use, emissions and economic growth, a less substantive focus on political strategies is not necessarily surprising. However, the separation of quantitative decoupling analyses and more qualitative investigations into the political barriers and potentials towards zero-carbon futures or reduction of energy and materials use may present a problem in itself because it prevents discussion of more effective and realistic strategies based on empirical analyses.

Discussion and conclusions
At least since the publication of the seminal 'Limits to growth' report (Meadows et al 1972), a debate is ongoing between scholars who hold that unlimited economic growth is impossible on a finite planet, and other scholars who believe that human ingenuity will eventually overcome all potential limitations to economic growth. The emergence of the notion of 'sustainable development' has suggested that economic development and respect for planetary boundaries (Steffen et al 2015), to use a modern word, can be reconciled. Claims that a decoupling of GDP from resource use and environmental pressures would be possible were already formulated very early on (United Nations 1987).
To contribute to this debate, we deliberately designed this pair of review articles broadly, as we aimed to incorporate a variety of indicators to comprehensively assess the use of biophysical resources (materials and energy) as well as a key class of outflows, namely GHG emissions (Jackson and Victor 2019). GHG emissions are dominated by CO 2 , i.e. the compound resulting from the combustion of most fuels that humans currently use, and hence a quantitatively dominant outflow of all dissipative use of materials (as opposed to stock-building materials such as concrete or steel; Krausmann et al 2018). This focus on social metabolism in its entirety (Haberl et al 2019) has shown that different patterns can be discerned by focusing on different aspects of resource use, and that the perspectives and results of communities looking at various aspects of resource use differ considerably.

Synthesis of insights into past decoupling
The large body of literature focused on the causal interrelations between energy and GDP uses econometric time-series and causality testing methods, for example Granger causality, but often shows little interest in the energy indicators analyzed or in actual thermodynamic basis of their hypotheses (see part I, Wiedenhofer et al 2020). While no robust conclusion can be drawn on the direction of causality, these studies show that energy and GDP are strongly related. Stern (2011) has argued that energy is an important factor of production, hence energy scarcity imposes restrictions on economic growth, which supports results from biophysical economics (Kümmel 2011). We found no evidence in the reviewed literature that would question this assertion.
The second group of articles (section 3.2) pays a lot of attention to the meaning of the energy indicators used. Many of the authors in this community come from energy analysis and regard themselves as analysts of 'biophysical economics' (Cleveland 1987, Hall et al 2001, Kümmel 2011. Their conviction is that energy use is a key factor of production (Ayres 2016), and that the quality of energy is hence crucial for assessing the role of energy in the economy (Hall et al 1986, Giampietro 2006, Haberl 2006. The main conclusions are that useful exergy and GDP are tightly coupled and that at the useful stage of energy use there is no evidence for relative decoupling. However, this does not mean that decoupling is not possible between primary energy and GDP, which is important because GHG emissions and extraction of energy resources are linked to primary energy, not useful exergy (Haberl 2006). The conclusion from this literature is that primary energy use can be decoupled from GDP only to the extent to which conversion efficiency from primary energy to useful exergy can be increased.
The review of social metabolism studies based on MEFA methods (Haberl et al 2004, Fischer-Kowalski et al 2011, Krausmann et al 2017a exemplifies the richness of measures of resource use and their different specific meanings (section 3.3). This community is well aware of the importance of a rich set of indicators, in particular of the difference between production-based and consumption-based accounts. This literature suggests that production-based relative decoupling is frequent, although countries exist in which use of physical resources grows faster than GDP. This seems to happen especially at early stages of the agrarian-industrial transition when large stocks of infrastructures and buildings are accumulated, as well as in export-oriented countries where production of raw materials and early processing stages are dominant. Absolute decoupling is rare and generally only occurs during periods of low GDP growth (Steinberger et al 2013). At the global level, only relative decoupling can be observed (Krausmann et al 2017b). In recent years several global multi-regional input-output models have been established which allow allocating extracted primary resources to final demand of any economy (Inomata and Owen 2014, Wiedmann and Lenzen 2018). Consumption-based analyses suggest that decoupling of production-based material flows is often contrasted by increases of material footprints that are similar to those of GDP (Giljum et al 2014a, Pothen andSchymura 2015, Wiedmann et al 2015).
Current trajectories of material and energy use, whether suggesting decoupling of resource use from economic growth or not, cannot be correctly interpreted without considering past material and energy flows on which they are also based. Current stagnation in per capita territorial/production-based resource use (Fishman et al 2016, Bleischwitz et al 2018a, for example, depends on past material flows  and entail a substantial legacy for the future (Krausmann et al 2017c). Since some materials enter the socio-economic system to be consumed for their energy content while others are for building up stock (manufactured capital) (Haas et al 2015), it may well be that different strategies are needed to observe, analyse, and set targets for decoupling material use of these two streams. Therefore, more insights can be expected by moving from studies of the decoupling of GDP from one resource or emission indicator to analysing interdependencies between GDP and multiple resources flows, respectively material stocks and resource or emission flows (Haberl et al 2017, Krausmann et al 2017c).
In recent years, a hypothesized S-shaped curve of material growth suggesting a notion of 'saturation' , i.e. a stable level of materials use, has gained prominence. In the MEFA community, the idea of saturation has recently attracted more attention than the EKC. This would imply sustenance of a stable, perhaps high, level of materials use coinciding with a continued growth of GDP and perhaps other socioeconomic indicators, in accordance with the 'steady state economy' discourse (Daly 1973, O'Neill 2015. However, so far, no consensus could be achieved on many important conceptual questions. It remains unclear whether saturation should be defined as country totals or per capita, whether consumption-or production-based flows (or material stocks) should be stabilized, and whether saturation should be achieved at the same level for all countries (Müller et al 2011, Pauliuk et al 2013, Chen and Graedel 2015, Fishman et al 2016, Cao et al 2017, Bleischwitz et al 2018a. Moreover, stabilization at a high level may fall short of achieving many sustainability and climate targets. The literature on CO 2 and other GHG emissions is large and growing fast . Most of the studies on territorial CO 2 use econometric methods, and many are based on the EKC framework (section 3.1). Empirical support for the existence of an EKC-type inverted U-shape of the relation between CO 2 emissions and GDP is seldom found . This also holds for total GHG emissions (section 3.4). Even when the data seem to suggest such a curve, the downward-bent part of the curve is usually too far in the future to be of use in reaching ambitious climate targets such as the Paris accord. The GHG emission literature reviewed in section 3.4 suggests a similar pattern as for material use: relative decoupling is the norm rather the exception, but cases of absolute decoupling are rare. A recent study, however, has identified and analyzed 18 'peak-and-decline' countries in which CO 2 emissions are falling in both territorial and consumption-based system boundaries (Le Quéré et al 2019). The study concludes that emissions in these 18 countries fell by a median −2.4%/year (25-75 percentile: −1.4% to −2.9%/year) over the period 2005-2015. Almost half of that reduction has been due to a decline in the share of fossil fuels in final energy use. A bit over one-third resulted from reductions of energy use. The study provides evidence that these reductions were a result of targeted policies to promote renewables and raise energy efficiency, but also profited from relatively low GDP growth rates between 1%-2%/year, which is similar to decoupling rates observed in MEFA studies (Steinberger et al 2013). It also noted that rates of CO 2 reduction achieved so far fell short from those required to comply with stringent CO 2 reduction targets as those implied by the Paris climate accord.

Current state of decoupling in the last decade
Because the analysis of the literature has yielded only limited aggregate insight into elasticities between GDP and resource/emission indicators due to the variety of measures used in the literature to describe (de)coupling, we summarize some information on the last decade in figure 2. Elasticities were calculated as OLS log regressions over 10 years using the formula log(resource/emission) = α + β log(GDP) + ε. A median elasticity of CO 2 of 0.4 in the higher income class (top panels in figure 2) means that for 1% of GDP growth, production-based CO 2 emissions grew by 0.4%. Elasticities below zero indicate absolute decoupling and elasticities >1 that resources/emissions grew faster than GDP. Results should be interpreted with caution in particular for those parts of figure 2 where data were only available for few countries (see sample sizes in blue font color). Median values of elasticities are close to one for most of the indicators in the low-income class, while they are often substantially lower than one for the higher income class. For the higher income class, elasticities of consumption-based (CB) indicators are highest for material use and substantially lower for CO 2 and GHG. For the lower income class, the highest median values are found for production-based emissions. Negative elasticities, indicating absolute decoupling, are most frequent for production-based GHG emission accounts and consumption-based TPES and CO 2 accounts for high income countries. For other indicators, instances of absolute decoupling also exist in the group of high-income countries, but are very rare for lower income countries. Thus, the results from our regression analysis over a 10-year timeframe are largely consistent with the main findings from our literature review.

Implications for future decoupling research and policies
What, then, are the conclusions for the prospects to achieve absolute decoupling in the future? The analyzed literature provides ample evidence that a continuation of past trends will not yield absolute reductions of resource use or GHG emissions. So far, environment and climate policies have at best achieved relative decoupling between GDP and resource use respectively GHG emissions (Kemp-Benedict 2018, Haberl et al 2019). Exceptions include a group of 18 countries that have reduced CO 2 emissions in the last decade (Le Quéré et al 2019), and a few national cases, most of which are due to specific circumstances that probably should not be generalized (e.g. when falling resource use stems from economic crises; Shao et al 2017). This observed absolute decoupling, however, falls short from the massive decoupling required to achieve agreed climate targets (Jackson and Victor 2016). Of course, rare occurrence of absolute decoupling in the past does not represent proof that it cannot become more common in the future-and perhaps intensifying the policies implemented in 18 peakand-decline countries could yield sufficient decoupling of GDP and GHG emissions to achieve climate targets. Even if rapid deployment of renewable energy could be achieved, however, the world's addiction to material resources would likely not wane, as harnessing renewables also requires substantial investments into large-scale buildings (e.g. hydropower plants), machinery (e.g. wind turbines, photovoltaic power plants) and infrastructures (e.g. expansion and reinforcement of electric transmission grids; Beylot et al 2019, Watari et al 2019).
In any case, meeting the goals of the Paris Agreement will require new and more effective policies than those deployed so far. These need to be based on absolute-not relative-reduction goals for GHG emissions, which could strongly benefit from curbing growth of resource use (Krausmann et al 2020). The IPCC 1.5 • C report (IPCC 2018) shows that even if high hopes are placed in future deployment of negative emission technologies, fast and deep cuts in global GHG emissions are required in order to address the 2.0 • C target agreed upon in the Paris climate accord, and even more so for reaching 1.5 • C. Currently, targets for reducing resource use or emissions are commonly framed as improvements of e.g. energy/GDP ratios. For example, SDG 7.3 aims at doubling the rate of energy intensity (energy/GDP) reduction, from approx. −1.5%/year to −3.0%/year. However, such targets allow substantial increases of resource use in absolute numbers if GDP growth is sufficiently fast (Heun and Brockway 2019). Hence, absolute GHG reduction goals can only be achieved if absolute goals for emission reductions are agreed upon. The analysis of policies and strategies (section 4) shows that decoupling research is so far poorly equipped to deal with this challenge. Only a tiny fraction of the decoupling literature in our random sample adopted a 'degrowth' perspective, which we have defined very broadly as a worldview allowing to question the priority of GDP growth over environmental goals. Whether one follows the viewpoint that a decoupling of GDP from environmental impacts is impossible (Ward et al 2016, Hickel andKallis 2019) may be less important than accepting the need to achieve absolute reductions of emissions regardless of GDP trajectories. Similar considerations apply to the use of many other biophysical resources (Green and Denniss 2018, Lazarus and van Asselt 2018).
A recent review suggest that strategies towards efficiency have to be complemented by those pushing sufficiency (Parrique et al 2019), that is, 'the direct downscaling of economic production in many sectors and parallel reduction of consumption' (p 3). Although concrete political strategies towards sufficiency-or degrowth-are still fragmented and diverse, they may include restrictive supply-side policy instruments targeting fossil fuels (instead of relative efficiency improvements), redistribution (of work and leisure, natural resources and wealth), a decentralization of the economy or new social security institutions (that complement the growthoriented welfare state). Recently suggested policies include moratoria on resource extraction and new infrastructures (e.g. coal power plants, highways, airports), bans on harmful activities (e.g. fracking, coal mining), the reduction of working hours and redistributive taxation, instead of just putting a price on resources and emissions (Schneider et al 2010, Kallis 2011, Koch 2013, Sekulova et al 2013, Jackson 2016, Green and Denniss 2018, Hickel and Kallis 2019. A new study suggests, however, that even energy sufficiency actions may be associated with rebound effects and negative spillovers (Sorrell et al 2020).
In any case, recent research suggests that states have so far refrained from strategies of sufficiency as these may contradict their claimed structural dependence on economic growth for the generation of tax revenue, employment and consumption-based political legitimacy. A strategic turn towards sufficiency that involves reductions in overall consumption levels and may lead to a degrowing economy might therefore pose a fundamental challenge to contemporary states-and liberal democracies (Pichler et al 2018, Hausknost 2019, Koch 2019). Studies in sustainable consumption increasingly argue that a decisive turn towards 'strong sustainable consumption governance' (Lorek and Fuchs 2013), that is, a clear focus on reducing the volume of the materials and energy resources consumed while maintaining levels of wellbeing, will be a key required for deep decarbonization.
Another recent strand of literature is focused on overcoming GDP as key target indicator of economic policy (Hoekstra 2019). This debate suggests that GDP may be becoming an increasingly irrelevant measure of welfare, as it was only loosely coupled with wellbeing in OECD countries over the last 40 years (Hoekstra 2019). In this view, GDP should be replaced or at least complemented by measures of wellbeing and planetary health, as suggested in the dashboard approach of the Sen-Stiglitz-Fitoussi-report (Stiglitz et al 2009), and in the Sustainable Development Goals. Scholars increasingly focus more on improving social wellbeing rather than GDP growth. One conceptual angle is the 'stock-flowservice' nexus approach (Haberl et al 2017(Haberl et al , 2019 suggesting that designing currently resource-intensive systems to provide for key contributions to social wellbeing (e.g. access/transport, housing/shelter, provision of food) in a resource-sparing manner in the first place can deliver these services at much lower levels of resource inputs than now. An example would be spatial patterns of settlements and work places that minimize the need for commuting, and foster commuting by environmentally friendly means such as walking, cycling or use of public transit. Such a focus on demand-side measures consistent with provision of services that are vital for social well-being is at the core of a currently emerging research community (Cullen et al 2011, Creutzig et al 2016, Brand-Correa and Steinberger 2017, Carmona et al 2017, Lamb and Steinberger 2017, Vita et al 2019. Perhaps the question to what extent GDP can be decoupled from resource use or emissions will turn out to be less important than the question how a good life for all on the planet can be organized within the planet's environmental limits (O'Neill et al 2018). Reductions in resource use and emissions commensurate with climate and sustainability goals (IPCC 2018, TWI2050 2018) may still be achieved by turning towards sufficiency and other transformative strategies. Anke Schaffartzik acknowledges financial support from the Spanish Ministry of Economy and Competitiveness, through the 'María de Maeztu' program for Units of Excellence (MDM-2015-0552). We gratefully acknowledge help in managing references by research assistants Andrea Gutson, Lisa Laßnig, Vivianne Rau, Anna Untersteiner, by Nicolas Roux for calculation of elasticities in figure 2, and the constructive comments of two anonymous reviewers.

Data availability statement
Any data that support the findings of this study are included within the article.